Upload
vodang
View
214
Download
0
Embed Size (px)
Citation preview
International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2013): 4.438
Volume 4 Issue 4, April 2015
www.ijsr.net Licensed Under Creative Commons Attribution CC BY
Application of Curve Fitting and Surface Fitting
Tools for High Leverage Points and Outliers of
Wind Speed Prediction
Prema V., Uma Rao K.
Dept. of EEE, R.V. College of Engineering, Mysore Road, Bangalore 560059, India
Abstract: Climate change is generally accepted as the greatest environmental challenge our world is facing today. Together with the
need to ensure long-term security of energy supply, it imposes an obligation on all of us to consider ways of reducing our carbon
footprint and sourcing more of our energy from renewable sources. Wind energy is one such source and this paper proposes a few
methodologies to predict the speed of wind. Multivariate regression model, curve fitting, surface fitting model and time series model are
implemented for the test data collected from a wind farm. The results obtained are then compared with the actual values available for
validation.
Keywords: multivariate regression; time series; wind speed; prediction; curve fitting; surface fitting.
1. Introduction
Intermittency of wind is the biggest challenge in
implementing wind-energy as a reliable autonomous source
of electric power. Energy crisis, global warming, depletion of
fossil fuels are the major factors looming the world today.
Adequate utilization of renewable energy sources like wind,
solar, biomass etc. proves to be the only alternative to
overcome these problems. Because of the indeterminate
nature and complementary behavior of wind and solar
energy, many researchers are putting their effort towards
hybrid energy systems. A hybrid energy system uses a
combination of sources like wind and solar along with a
battery and Diesel Generator set.
Optimal allocation of the available resources is one of the
challenges faced in the design of hybrid energy systems. For
a well-planned system, it is essential to have accurate
predictive models of the wind and solar sources. Wind
energy is directly dependent on the wind speed available at
that location and this speed is extremely erratic. Hence, a
model is needed for accurate prediction of wind speed. This
paper proposes a few predictive models for wind energy
using regression and curve fitting.
A wide variety of methods were proposed in the literature for
wind forecast [1]. Persistence method or the „Naïve predictor
is the simplest and fundamental prediction model. Physical
approach makes use of Numerical Weather Predictor model
such as global forecasting system; prediktor etc. Statistical
approaches include time series forecasting methods and
Artificial Neural Network models. A few hybrid models are
also proposed which used a combination of the above
mentioned methods. This paper proposes four different
models for wind forecast based on regression, curve fitting
and surface fitting.
The correlation between the physical parameters was
obtained by using the statistical data analysis tools. Linear
regression, nonlinear regression, curve fitting and surface
fitting models were developed. A time series model was also
implemented for the purpose of tracking the peak overshoots
and undershoots. Here, time was defined as a variable along
with temperature, pressure and relative humidity and the
previous value of the wind speed was used as the input.
Errors for the various models were also calculated.
2. Parameters Used
There are various variables on which wind speed depends
upon and this paper presents the statistical relationship
among these variables which impact the wind speed in a wind
farm in Bagalkote region in Karnataka using MATLAB. The
parameters are typical of any moderate tropical country. The
variables used are temperature, relative humidity and
atmospheric pressure.
2.1. Atmospheric Pressure (P)
It is the force per unit area exerted on a surface by the weight
of air above that surface in the atmosphere of Earth. In most
circumstances atmospheric pressure is closely approximated
by the hydrostatic pressure caused by the weight of air above
the measurement point. On a given plane, low-pressure areas
have less atmospheric mass above their location, whereas
high-pressure areas have more atmospheric mass above their
location. Likewise, as elevation increases, there is less
overlying atmospheric mass, so that atmospheric pressure
decreases with increasing altitude. The unit of atmospheric
pressure used in this paper is consistent throughout and it is
mm of Hg.
2.2. Relative Humidity (RH)
It is the ratio of the partial pressure of water vapor in an air-
water mixture to the saturated vapor pressure of water at a
prescribed temperature. The relative humidity of air depends
on temperature and the pressure of the system of interest.
There is a considerable degree of difference between the
relative humidity measured indoors and that measured
Paper ID: SUB153600 2585
International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2013): 4.438
Volume 4 Issue 4, April 2015
www.ijsr.net Licensed Under Creative Commons Attribution CC BY
outdoors in regions such as wind farms and those at altitudes
of the shaft locations. Relative humidity, being a ratio has no
unit for and the numbers used in this paper are expressed as a
percentage.
2.3. Temperature (T)
A temperature is a numerical measure of hot and cold. Its
measurement is by detection of heat radiation or particle
velocity or kinetic energy, or by the bulk behaviour of
a thermometric material. The kinetic theory indicates
the absolute temperature as proportional to the average
kinetic energy of the random microscopic motions of their
constituent microscopic particles such as electrons, atoms,
and molecules. The basic unit of temperature in
the International System of Units (SI) is the Kelvin. For
everyday applications, it is often convenient to use the
Celsius scale, in which 0°C corresponds very closely to
the freezing point of water and 100°C is its boiling point at
sea level.
2.4. Wind Speed or Wind Velocity (Ws)
Wind is caused by air moving from an area of high pressure
to low pressure. Wind speed affects weather forecasting,
aircraft and maritime operations, construction projects,
growth and metabolism rate of many plant species, and
countless other implications. It is now commonly measured
with an anemometer but can also be classified using the older
Beaufort scale which is based on people's observation of
specifically defined wind effects. The unit of wind speed
used in this paper is consistently meters per second.
3. Data Acquisition
The data used for the proposed work was collected from a
wind farm in Bagalkot, Karnataka. The recorded data
consists of values of Air Temperature, Relative Humidity,
pressure, wind speed and the wind direction measured for
different altitudes, for every minute over a period of six
months. For analysis, one particular altitude values were
considered. The choice is made based on the fact that wind
velocity is higher in upper strata of atmosphere thereby
having more potential to generate wind energy[6]
. The data in
Microsoft Excel format is imported to MATLAB for the
analysis or modelling.
4. Modelling for Wind Speed Prediction
In regression models, the forecast is expressed as a function
of certain parameters which influence the outcome.
Regression gives an explanatory model that relates output to
various inputs which facilitates a better understanding of the
situation and different combinations of inputs can be tested
and their effects on the forecast can be studied. The
forecasting parameter is the wind speed and the independent
parameters used are atmospheric pressure, relative humidity
and average temperature.
Linear and non-linear regression models are developed.
Linear regression model is implemented by restricting the
number of dependent variables to one, which is wind speed.
Multivariate linear regression is often found to be more
efficient and useful since it considers the effects of all the
parameters that may have a significant effect on the
dependent variable present. In non-linear regression, the
inputs or the independent variables can have a non-linear
relation with the wind speed.
Curve fitting tool in MATLAB is used to find the best fit
between wind speed and one of the input variables. The best
fit is chosen to develop the model. Surface fitting is another
technique used to find the best fit between wind speed and
two of the input variables. A similar model was developed. A
time series model is also developed which takes the previous
value of wind speed and time as the inputs.
4.1. Linear Regression model
The input parameters considered for the linear regression are
combinations of the measured quantities (Pressure,
Temperature and Relative Humidity). A number of models
have been tested and the best model has been chosen which is
shown in equation 1. The inputs of temperature, pressure,
relative humidity, product of temperature and relative
humidity, product of temperature and pressure were observed
to yield the best results. The plot is shown in Figure 1 WS= -30.079*T+ 0.1257*RH- 0.0353P- 0.006*T*RH + .044*T*P (1)
4.2. Non-linear Regression model
A non-linear relation between the various combinations of
inputs was modelled using regression tool. It was found from
the analysis of theoretical data that there is proportionality
between wind speed and square root of pressure. This
relationship was also considered in the model. Various
combinations were tried out and the best of the lot was
chosen. The model is shown in equation 2. The results are
shown in fig.2.
WS= 0.3143*T+ 0.0408*RH - 0.2633*sqrt(P) (2)
4.3. Curve Fitting model
To implement this model, first the relationship between wind
speed and pressure is derived. This is obtained by plotting
the scatter diagram between pressure and wind speed. The
best fit is plotted as shown in fig.3. There is additional option
for including weights in the data points. The model thus
obtained is shown in equation 3.
WS = -9.565e-010 *(P)^3.757+53.08 (3)
4.4. Surface Fitting Model
The inputs yielding best results were those of pressure and
relative humidity and the surface fit is as shown in Figure 5.
The output graph obtained was slightly different from that
obtained from the previous functions and is given by Figure
6. The difference noted was the increased correlation
between the inputs and lower values of wind speed. The
obtained model is given in equation 4.
WS= -1525 +13.35*RH+4.131*P+ 0.00332*(RH)2 -
0.01909*RH*P -0.00278*(P) (4)
Paper ID: SUB153600 2586
International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2013): 4.438
Volume 4 Issue 4, April 2015
www.ijsr.net Licensed Under Creative Commons Attribution CC BY
Figure 1: Plot for linear regression model
Figure 2: Plot for non-linear regression model
Figure 4: Plot for curve fitting model
Figure 3: Scatter diagram for curve fitting
Figure 5: Scatter diagram for surface fittng
Figure 6: Plot for surface fitting model
Paper ID: SUB153600 2587
International Journal of Science and Research (IJSR) ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2013): 4.438
Volume 4 Issue 4, April 2015
www.ijsr.net Licensed Under Creative Commons Attribution CC BY
4.5. Time Series Model
Even in the best fit, the generated curve matched the original
curve in phase alone but not in magnitude. Observing all the
results, it can be concluded that an open loop system will
never give the required correlation. Hence, feedback systems
were incorporated in order to account for the mismatched
magnitudes. Thus, time series models were implemented to
improve the response.
In time series modelling, time was defined as a new variable
along with temperature, pressure and relative humidity. The
present value was predicted with the help of previous value
and the error. The input vector „X‟ was defined for an
intercept along with temperature and pressure.
The output curve showed better correlation with the input.
But, the output had very large offshoots causing large errors.
The output was enhanced by providing error correction by
shifting the output waveform by trial and error. Figure 7
shows plot. The following is the equation obtained after
modelling using time series is given in eq: 5
WS(time+1) = WS(time) + 0.2018 * T - 2.0873 (5)
Figure 7: Plot for time series model
5. Results and Discussions
As seen, the predicted values form a graph that is following
the trend of the plot formed by the actual values. The ups and
downs are followed, but the numerical data differs in the case
of multivariate regression models. In the case of time series
model both trend and the magnitude matched to a greater
extent.
Table 1: Comparison of Errors
Model SSE
Model regress 3763.851
Model nlinfit 3796.608
Model cftool 4178.766
Model sftool 3746.625
Time series model 3568.12
Table 1 shows the values of sum of squared errors for the
various models implemented.
6. Conclusion
Five different models were implemented for the prediction of
wind speed. It can be observed from the results and
calculated errors that the time series model has the least sum
of squared errors taken for about 942 points. Hence the
model with feedback elements yielded the best result. The
time series element solved the magnitude mismatch problem
and predicted wind speed with the highest accuracy as can be
seen in the output.
References
[1] Saurabh S. Soman, Hamidreza Zareipour, Om Malik, and
Paras Mandal, “A Review of Wind Power and Wind
Speed Forecasting Methods With Different Time
Horizons”, North American Power Symposium (NAPS),
2010 pp 1- 8
[2] A. M. Foley, P. G. Leahy, A. Marvuglia, and E. J.
McKeogh, "Current methods and advances in forecasting
of wind power generation" Renewable Energy, vol. 37,
pp. 1-8, Jan 2012.
[3] Chatterjee, S., and A. S. Hadi. "Influential Observations,
High Leverage Points, and Outliers in Linear
Regression”, Statistical Science. Vol. 1, 1986, pp. 379–
416.
[4] Seber, G. A. F., and C. J. Wild. “Nonlinear Regression ".
Hoboken, NJ: Wiley-Interscience, 2003.
[5] DuMouchel, W. H., and F. L. O'Brien. “Integrating a
Robust Option into a Multiple Regression Computing
Environment." Computer Science and Statistics:
Proceedings of the21st Symposium on the Interface.
Alexandria, VA: American Statistical Association, 1989.
[6] Holland, P. W., and R. E. Welsch. "Robust Regression
Using Iteratively Reweighted Least-
Squares."Communications in Statistics: Theory and
Methods, A6, 1977, pp. 813–827
Author Profile Prema V. received her Masters degree in Power
Electronics from Visvesvaraya Technological
University in 2005 and Bachelor‟s degree in Electrical
& Electronics Engineering from Calicut University in
2001. She is currently working as Assistant Professor
in R.V.College of Engineering, Bangalore, India. Her research
interests include Power Electronics, Renewable Energy and Digital
Signal Processing.
Dr. Uma Rao K. was born on 10th March 1963. She
obtained her B.E in Electrical Engineering and M.E in
Power systems from University Visvesvaraya college
of Engineering, Bangalore and her Ph.D from Indian
Institute of Science, Bangalore. She is currently Professor,
department of Electrical & Electronics Engineering, R.V. College
of Engineering, Bangalore. Her research interests include FACTS,
Custom Power, Power Quality , renewable energy and technical
education.
Paper ID: SUB153600 2588